Gladia

Solaria-3 is out. And it's the best speech model for European business audio today. We've spent months testing speech models on our own customers' audio. The gap between benchmark scores and production performance is brutal. Models claiming 3% WER on clean audio regularly hit 15%+ on a real sales call with a non-native speaker. So we decided to build differently. Solaria-3 was tested on real customer calls, annotated by humans — not just on curated benchmarks. And it's our most accurate model yet across all five European languages, most popular today among our users: English, French, German, Spanish, and Italian. Results are impressive: → #1 on English production audio (9.6% WER — 26% better than Solaria-1) → #1 on Earnings22, the industry standard for business and financial speech → #1 on Switchboard — the hardest conversational telephone benchmark. ElevenLabs hits 55.2% WER. We hit 33.9%. → Near-best on noisy audio — 1.4% WER → Consistent improvement across EN, FR, DE, ES, IT on real customer audio. We also publish all numbers, including the two benchmarks where we step back vs our previous model, which will remain available for broader language coverage. On the landing page, you’ll also find real transcripts where Solaria-3 was the only model to capture all the info correctly, including with challenging business terms and accents. What an innovation leap with this model! Huge congrats to the team. Oh, the best part: it’s free for everyone to try! Link in comments.

We just shipped the first-ever market map of the meeting assistant space, featuring Granola, Fathom, Fireflies, Gong, and many more. Meeting assistants are no longer just productivity tools. The best ones are racing to become the central nervous system of how teams collaborate, decide, and act. And in a market moving this fast, knowing where each player is placing their bets — and where the real moats are — matters. A year of research, a labor of love, with lots of valuable inputs: - Analysis of 100+ tools reviewed and placed half across four moats - A global survey of 2,000+ people actively using meeting assistants - Interviews with investors from Northzone and Sequoia - Conversations with founders from Nabla, Recall, and more

We just shipped the first-ever market map of the meeting assistant space, featuring Granola, Fathom, Fireflies, Gong, and many more. Meeting assistants are no longer just productivity tools. The best ones are racing to become the central nervous system of how teams collaborate, decide, and act. And in a market moving this fast, knowing where each player is placing their bets — and where the real moats are — matters. A year of research, a labor of love, with lots of valuable inputs: - Analysis of 100+ tools reviewed and placed half across four moats - A global survey of 2,000+ people actively using meeting assistants - Interviews with investors from Northzone and Sequoia - Conversations with founders from Nabla, Recall, and more

Today we’re releasing something I've been dreaming about for over a year: the first-ever market map for meeting assistants. This category has moved faster than I ever could have imagined. What started as simple recorders has evolved – in less than 4 years – into a race to build the next indispensable layer of work. Deeper integrations. Wider context capture. Proactive, on-call agents that don't wait to be asked. The innovation is remarkable. But the challenges are real. User retention is still the category's open wound. Most tools look interchangeable to the people using them. Loyalty is fragile. The Slack or Figma moment, where one player becomes truly indispensable for workplace collaboration, hasn't happened yet. We've had a front-row seat to all of it. These teams are our earliest customers, and we've been in the trenches with them since the beginning, powering STT while they figured out how to stand out in this space. So we decided to make something for them. A comprehensive market map. 100+ tools reviewed. 2,000+ end users surveyed across six countries – about their frustrations, habits and unmet needs. Conversations with visionary VCs from Sequoia and Northzone. And interviews with the founders themselves. The result: in-depth market insights, with four quadrants mapping where key players sit across capture, intelligence, integrations, and market strategy. A clear read, for any vendor in this space, on where the market is heading, and where the real opportunities to win are. And we're giving it away. Because this is what we owe the people who bet on us early. Not a sales pitch. A love letter to every team that has been building something genuinely hard, and doing it brilliantly. To the founders, PMs and investors navigating this market: you rock 💜 To the marketing team at Gladia: this report is pure magique. Beyond proud. 📒 Check it out: https://lnkd.in/deEeGNeF

Benchmarks in speech AI are broken. Everyone knows it. Nobody talks about it. When I joined Gladia as CTO, one thing quickly became painfully obvious: - No one could reproduce speech benchmarks. - Two teams could test the same model and get completely different results. So, the team and I went digging and we realized that normalization was a big part of the explanation. As a result, we open-sourced the whole methodology. Normalization rules, scripts, all of it. Reproduce our results. Challenge them. Watch to see what we found 😏 Full repo + evaluation pipeline in comments.

We’re open-sourcing our entire benchmarking methodology. In the past, I vented a lot about benchmark chaos in the speech industry. Today, I want to introduce the person who helped me solve it - and share our entire repo with you. A few months ago, Maxime joined Gladia as our new CTO and immediately took this problem head-on, with a smile. He turned my frustration into an engineering problem to solve. Normalization rules. Scoring scripts. Reproducibility. And today, we’re open-sourcing the whole benchmarking methodology. Because everyone deserves transparency about this. Watch Max explain it! A huge step for me, and the team that, I hope, will benefit everyone - founders, customers, and the community. Take it, play with it, challenge it.